01a Introduction to Biostatistics

Julius-Maximilians-University Würzburg
Course: “Biostatistics”
Translational Neuroscience

Dr. Mario Reutter (slides adapted from Dr. Lea Hildebrandt)

Welcome!

Hello!

Who am I?

Who are you?

• What is your background?

• Do you have experience with data analysis?

• What’s your attitude towards statistics?

Using the Slides

• These slides are created directly in R with the quarto extension.
• You can jump to a slide by clicking the three dashes in the bottom left.
• You can conveniently copy R code from the slides with one click and paste it into your RStudio.

print("Hello World")

[1] "Hello World"

Organizational Issues

• Attendance is mandatory! (I don’t make the rules :) )

  • You may miss one session without giving reasons (recommendation: don’t waste it early!)

  • If you miss additional sessions, please write an email with an explanation (further proof may be required, e.g. doctor’s certificate)

• The course will take place in person

  • Only for students who are not in Würzburg yet, there is an option via Zoom

• We will use these textbooks (Open Educational Resources - freely available online, also linked in WueCampus):

  • Statistical Thinking for the 21st Century

  • For the R part: Fundamentals of Quantitative Analysis

  • (Additional Resource: R for Data Science)

Contents

• From basic probability to (Generalized) Linear Mixed Models
  • Some things may be repetitive for you but this course aims to provide a common starting position for your next semesters
• Input (lecture style) with hands-on R sessions
• In addition, you should read a few pages in the text books
  (Statistical Thinking for the 21st Century, and possibly Fundamentals of Quantitative Analysis)
• Project: Independently analyze a dataset (exam with pass or fail grading)

For some, e.g. the psychologists, it will be more of a repetition - but you will also learn R.

Slides might be text heavy → so that you can go through the slides afterwards again (but textbook might also be helpful)

Project

• Find a dataset that can answer a question you are interested in

  • https://ourworldindata.org/
  • Statistisches Bundesamt
  • Your own, e.g., from an internship

• preprocess/wrangle it,

• analyze the data,

• and write a short (min. 2-page) report!

  • short intro incl. research question and hypothesis
  • methods (both how the data were acquired and how they are analyzed)
  • results (incl. at least one plot)
  • and a short discussion.
    • All these parts should be at least half a page long.

Schedule

The up-to-date version can be found at https://spressi.github.io/biostats/

Date	Topic	Reading	Lecturer	Project Deadlines
16.10.	General & R Intro	ST21: 1-3, QF: 1-3	Reutter	R & RStudio installed
23.10.	Probability	ST21: 4, QF: 4-6	Hildebrandt
30.10.	Data Wrangling	ST21: 4, QF: 4-6	Reutter
06.11.	Data Visualization	QF: 7	Reutter
13.11.	Sampling	ST21: 7-8, QF: 8	Hildebrandt	Dataset
20.11.	Penance Day
27.11.	Probability & Sampling in R	ST21: 7-8, QF: 8	Hildebrandt
04.12.	Hypothesis Testing	ST21: 9-10	Reutter
11.12.	Comparing Means & Categories	ST21: 12, 15	Reutter	Research Question & Hypotheses
18.12.	Exercises (t-Tests, Chi²)	ST21: 12, 15	Reutter
25.12. & 01.01.	Christmas & New Year’s
08.01.	(General) Linear Models	ST21: 12-13	Hildebrandt	First Analysis Ideas
15.01.	Exercises (GLM)	ST21: 12-13	Hildebrandt
22.01.	Linear Mixed Models	ST21: 14	Hildebrandt	Analysis (with R scripts)
29.01.	R Markdown for Reports		Reutter
05.02.	Troubleshooting Your Report		both	Send Questions via Email
12.02.	Reproducible Research	ST21: 18	Hildebrandt	Report (with R Markdown)

ST21: Statistical Thinking for the 21st Century
QF: Fundamentals of Quantitative Analysis (QuantFun)

Why Statistics?

Why is it important that YOU know statistics?

• You’re doing a research master!

  • Research = Reading & understanding papers (esp. the analyses)
  • Designing your own experiments, analyze data, interpret results

• We live in an increasingly data-centric world

  • Knowing how to wrangle and analyze data is a valuable skill

• Facts & data literacy matter more than ever!

  • Fake News, “Lying with stats”, Reproducibility Crisis

  • Being able to call bullshit (https://www.callingbullshit.org/)

• “I only believe in statistics that I doctored myself” ― Winston S. Churchill

  • However: “It is easy to lie with statistics, but easier to lie without them” ― Frederick Mosteller

What is Statistical Thinking?

• “a systematic way of thinking about how we describe the world and use data [to] make decisions and predictions, all in the context of the inherent uncertainty that exists in the real world.” (Poldrack, Preface of ST21)

• “Statistical thinking is a way of understanding a complex world by describing it in relatively simple terms that nonetheless capture essential aspects of its structure or function, and that also provide us some idea of how uncertain we are about that knowledge.” (Poldrack, Chapter 1)

break down complexity, include uncertainty

Why is Statistical Thinking Important?

• data literacy vs. intuition/heuristics/anecdotal evidence
  • Public discourse about Covid-19, migration, etc. (e.g., “50% of people in intensive care are vaccinated”)

Base Rate Fallacy

example availability heuristic from book (or any other example where intuition is wrong, i.e. vaccinations/covid…)

What can Statistics Do For Us?

• Describe patterns by summarizing/breaking down data (“descriptive statistics”)

• Decide whether one thing is better than another, given the uncertainty (“inferential statistics”)

• Predict how other people would “behave” (generalize to new observations)

describe: not useful to look at every single data point/person, but we need s.th. like tendencies/trends…

The Big Ideas

• Learning from data: Update our beliefs

• Aggregation: How to summarize the data to draw meaningful conclusions?

• Uncertainty: Probabilistic evidence

• Sampling from the population: Which people etc. do we select?

ask for every point what I could mean w/ it?

Learning from data: gather new knowledge or even just hypotheses

Aggregation: Can’t look at all ind data points, need to find trends etc. (should not go too far! throwing out data)

Uncertainty: stats = tools for making decisions under uncertainty, we can never prove anything but provide evidence, there is no 100% certainty for an outcome

sampling: how do we represent the population? What is the population? how much data do we need? More is better, but payoff decreases…

Causality

Correlation does not imply causation… but is a hint!

Example: Smoking = less risk for Parkinson’s disease? (Godwin-Austen et al., 1982; Chen et al., 2010)

→ confounding factors?

e.g., individual dopaminergic activity ⇒ addiction & motor function

Randomized Controlled Trials (RCT) as the solution?

RCT: exp control and manipulation, removes confounds if done well
At least some more causal evidence!

QUESTIONS so far?

What are Data?

• What do you think are data?

• qualitative vs. quantitative

  • qualitative?

    • open questions, descriptions… can potentially be coded into categories

  • quantitative?

    • numeric, can be averaged etc.

Chat → after showing slide! Come up with examples for “Data”

Collect: Do you have ideas? What are data you encounter in your lives/work etc? What are differences between these data?

What are Data? (2)

• Data types

  • character/string: text (qualitative)

  • factors/categories

  • types of numbers (quantitative)

    • binary: 0 or 1, TRUE or FALSE (logical)

    • integers: whole numbers

    • real numbers: decimals/fractions

• discrete vs. continuous

  • discrete: finite set of particular values (0 or 1, scale from 1 to 10)

  • continuous: real numbers that fall into particular range (e.g., brain activity, visual analoge scale)

• What data type is eye color?

eye color can be categorical (e.g., “brown”, “blue”, “green”) or numerical (wave length in nm)

What is a Data Set?

• a collection of data

• usually organized into rows and columns (like an excel spreadsheet)

  • rows: participants/animals/cells…

  • columns: variables!

    • each variable contains one type of measurement

  • table cells = unique observations of variables per participant etc.

NHANES dataset

possibly go through columns and ask for data types?

What Makes a Good Measurement?

• What is being measured?

  • constructs vs. proxies: need to be well-defined! (Difficult)

  • measurement error

    • random: e.g., variation in reaction times of same participant across trials

    • systematic: e.g., miscalibrated eye-tracking device

• Do we have a “gold standard” to compare the measurement to?

Break-Out session: Brainstorm what makes a good vs. bad measurement!

Group work/brainstorm:

• What are problems?

• Which kind of errors/when is data NOT good

• how can we minimize error?

Reliability

Correlation of a measurement with “itself”

• Internal reliability (consistency)

• Test-retest reliability (stability)

• Inter-rater reliability (agreement)

Correlation with other variables can’t be higher than reliability (cf., Wilmer et al., 2012, Table 1)!

Validity

Are we measuring the construct we’re interested in?

• Face validity: Does it intuitively make sense? First reality check!

• Construct validity

  • convergent validity: Related to similar measures that should measure the same construct

  • divergent validity: Is it unrelated to other measures?

• Predictive validity: Is it predictive of other outcomes? (e.g., intelligence & job success)

Validity & Reliability

Reliability & Validity

Summarizing Data

• Throwing away (some of the) information!

  • extract the quintessence of the data (important for forming models)

  • make predictions

• Counts, frequencies, percentages, averages

Thanks!

Learning objective today:

• Why is statistics important?

• What are data?

• What is reliability and validity?

Next:

• R session: “Intro to R”

  • please install R & RStudio!

    • Installation tutorial: https://learningstatisticswithr-bookdown.netlify.app/intror

    • You can also already install Rtools, which will be needed later: https://cran.r-project.org/bin/windows/Rtools/

• Theory: Models